mirror of
https://github.com/oobabooga/text-generation-webui.git
synced 2024-11-25 01:09:22 +01:00
61 lines
2.0 KiB
Python
61 lines
2.0 KiB
Python
'''
|
|
|
|
Converts a transformers model to a format compatible with flexgen.
|
|
|
|
'''
|
|
|
|
import argparse
|
|
import os
|
|
from pathlib import Path
|
|
|
|
import numpy as np
|
|
import torch
|
|
from tqdm import tqdm
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
parser = argparse.ArgumentParser(formatter_class=lambda prog: argparse.HelpFormatter(prog,max_help_position=54))
|
|
parser.add_argument('MODEL', type=str, default=None, nargs='?', help="Path to the input model.")
|
|
args = parser.parse_args()
|
|
|
|
def disable_torch_init():
|
|
"""
|
|
Disable the redundant torch default initialization to accelerate model creation.
|
|
"""
|
|
import torch
|
|
global torch_linear_init_backup
|
|
global torch_layer_norm_init_backup
|
|
|
|
torch_linear_init_backup = torch.nn.Linear.reset_parameters
|
|
setattr(torch.nn.Linear, "reset_parameters", lambda self: None)
|
|
|
|
torch_layer_norm_init_backup = torch.nn.LayerNorm.reset_parameters
|
|
setattr(torch.nn.LayerNorm, "reset_parameters", lambda self: None)
|
|
|
|
def restore_torch_init():
|
|
"""Rollback the change made by disable_torch_init."""
|
|
import torch
|
|
setattr(torch.nn.Linear, "reset_parameters", torch_linear_init_backup)
|
|
setattr(torch.nn.LayerNorm, "reset_parameters", torch_layer_norm_init_backup)
|
|
|
|
if __name__ == '__main__':
|
|
path = Path(args.MODEL)
|
|
model_name = path.name
|
|
|
|
print(f"Loading {model_name}...")
|
|
#disable_torch_init()
|
|
model = AutoModelForCausalLM.from_pretrained(path, torch_dtype=torch.float16, low_cpu_mem_usage=True)
|
|
#restore_torch_init()
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(path)
|
|
|
|
out_folder = Path(f"models/{model_name}-np")
|
|
if not Path(out_folder).exists():
|
|
os.mkdir(out_folder)
|
|
|
|
print(f"Saving the converted model to {out_folder}...")
|
|
for name, param in tqdm(list(model.model.named_parameters())):
|
|
name = name.replace("decoder.final_layer_norm", "decoder.layer_norm")
|
|
param_path = os.path.join(out_folder, name)
|
|
with open(param_path, "wb") as f:
|
|
np.save(f, param.cpu().detach().numpy())
|